Back to Gap Map Infrastructure / Classic ML & computer vision

Depth Pro

Apple
restricted / Overall score: 2.8

Depth Pro is Apple's monocular depth model: from a single image it produces a metric depth map, with absolute scale, and estimates the camera's focal length without needing camera metadata. A multi-scale Vision Transformer built on DINOv2 encoders does the work, and the repository's model is a re-trained reference implementation close to, but not matching, the paper's. A Transformers-format checkpoint is also on the Hugging Face Hub.

Openness

2 high confidence
2.0
weights
open (depth_pro.pt and the Transformers checkpoint on the Hub, ungated)
data
documented-not-released (the card lists the real and synthetic training datasets in a table; no mixture is released)
code
partial (inference, command-line and boundary-metric evaluation code; no training code)
license
Apple-AMLR (Apple Machine Learning Research Model license on the checkpoints; non-commercial research only. The repository code is under Apple's separate sample-code license)

The Depth Pro checkpoints carry Apple's research-model license, which limits use to non-commercial scientific research and excludes any commercial product or service. The repository code is under a separate permissive Apple license, but it covers inference and evaluation, not training.

Adoption

2 high confidence
2.0

Hugging Face downloads over the trailing 30 days for the two declared checkpoints, almost all of them for the Transformers version.

Capability

4 high confidence
4.0

Depth Pro returns a metric depth map for any image the user supplies, without training, as Depth Anything does. It works from one image at a time, where Depth Anything 3 also takes several views and estimates camera pose.

Verified 2026-11-12